Yahoo Finance's Deirdre Bosa discusses an Anthropic (ANTH.PVT) claim that its Mythos model got Claude running on AMD (AMD) chips in about three days, raising questions about whether AI can erode the value of Nvidia's (NVDA) CUDA moat.
Nvidia is having a brutal November, you might have heard. A lot of the attention is on Jensen Wong. His board has opened an independent investigation, shareholders are suing lawmakers, they want answers. Um there's also this chart making the rounds in tech circles. Now it comes from Anthropic's IPO road show.
Anthropic says its Mythos model helped get part of another cloud model running on AMD chips in about three days. Now the company says that work would traditionally take 6 to 12 months. Have a listen.
This has also made us really good at bringing up new chips. As an example, this year we started working with AMD GPUs for the first time. In the past, bringing on a new chip platform took dozens of engineers months of work. This time, we had Mythos try it out. It got Claude running on AMD in just three days with basically no human intervention. Claude ported or wrote almost 100% of the kernels.
Okay, so why does this matter for Nvidia? Because this is a really hard problem to solve and a big part of the answer here is what it says about Cuda. That's the software platform that developers use to put Nvidia's chips to work. It is part of this Nvidia flywheel. Cuda is a reason to work on Nvidia GPUs and companies have really spent years building their software around it.
So switching to someone else's chips, that can mean months of work to get everything running properly. That gives customers a reason to stay within the Nvidia ecosystem, buy GPUs, use Cuda. It also helps Nvidia defend its prices.
But now that Anthropic is saying AI can help do that work much, much faster, that makes Nvidia, it makes Cuda, it makes Nvidia as a whole more vulnerable at really a key moment for the company. There's an important detail here as well. Um those three days covered one part of the process for one model. So getting it to read and process a prompt at a useful level of performance.
So there is that caveat, but still you can see why this has people talking. If AI makes it easier to move beyond Cuda, Nvidia's customers, they get more choice. This is sort of the big question of the last few years. That's why you're seeing all of these custom ASICs from Google, from Amazon, from Microsoft, from Open AI, um make runs sort of at the king at Nvidia over the last few years.
And the question for Jensen now becomes, how much extra will they pay to actually stay within that ecosystem if those walls are coming down a little.
